Predictive coding
Also known as: Predictive processing, Prediction-error minimization
The brain perceives by predicting the world and correcting only its errors.
What it means
Predictive coding is an influential theory of brain function in which perception is not a bottom-up readout of the senses but a continuous process of the brain generating top-down predictions about incoming signals and updating only on the basis of prediction error — the mismatch between what was expected and what arrived. On this view, higher levels of a hierarchy send predictions downward while lower levels send back only the residual error, an efficient code that transmits surprise rather than raw data. What we consciously perceive is thus the brain's best hypothesis about the causes of sensory input, with priors filling in and sometimes overriding the actual signal. The framework, often cast in Bayesian terms and generalized as free-energy minimization, unifies perception, attention (as the precision-weighting of errors), action, and learning, and it explains illusions, hallucinations, and the influence of expectation on what is seen. It matters as a candidate grand theory linking neuroscience, perception, and machine learning, though critics question its testability and scope.
Examples
We read a sentence with a missing or misspelled word without noticing, because the brain predicts the expected word and downweights the conflicting sensory evidence.
Lyrics you have misheard for years snap into place the moment someone tells you the real words; the sound never changed, only the prediction your brain brought to it.
A stair that isn't there jolts you badly — your brain confidently predicted solid ground at that height, and the whole mismatch arrives at once as error.
First described in Rao & Ballard (1999); developed by Karl Friston (free-energy principle).